01
What happened
The paper outlines a possible two-axis risk framework and discusses premarket evaluation through non-clinical benchmarking and clinical confirmation, along with risk-proportionate postmarket monitoring. It also asks about foundation models and agentic systems. FDA invited manufacturers, clinicians, consumers, researchers, the public, and other interested parties to submit comments by 19 October 2026.
Sources for this section: [1]
02
Why it matters
Generative systems can change outputs with prompts and context, inherit behavior from foundation models, and operate inside workflows where human oversight varies. Regulation must therefore connect intended use, failure severity, data and model change, validation population, cybersecurity, human factors, performance drift, complaint handling, and monitoring after deployment rather than relying on one benchmark score.
Sources for this section: [1]
03
What to watch next
Watch the public docket, subsequent FDA summaries, guidance or rulemaking, and how proposals distinguish locked functions from adaptive or agentic behavior. Also watch requirements for transparency, real-world monitoring, updates, responsibility across developers and health systems, and performance across demographic and clinical groups. Until then, the discussion paper should be described as a proposal under consultation.
Sources for this section: [1]
Sources
- FDA Seeks Public Feedback to Inform Regulatory Approach for Generative AI-Enabled Medical DevicesU.S. Food and Drug Administration · 2026